SLIDE 1
Bike share traffic predictions
using machine learning
Arnab Kumar Datta
Agenda
- Introduction to bike-sharing
- Motivation and vision
- A short introduction to
machine learning
- Overview of software
- Results
- Conclusion
Bike share traffic predictions using machine learning Arnab Kumar - - PowerPoint PPT Presentation
Bike share traffic predictions using machine learning Arnab Kumar - - PowerPoint PPT Presentation
Bike share traffic predictions using machine learning Arnab Kumar Datta Agenda Introduction to bike-sharing Motivation and vision A short introduction to machine learning Overview of software Results Conclusion
SLIDE 2
SLIDE 3
Bike-sharing
SLIDE 4
Why bike-sharing
SLIDE 5
The problem
SLIDE 6
Above: A customer reviews London’s bike-share system on the tripadvisor website
SLIDE 7
Above: A customer reviews Washington’s bike-share system on the tripadvisor website
SLIDE 8
Users currently have real-time systems
SLIDE 9
The vision
“I will be downtown at 8 am on Monday. Will the bike station be full?”
SLIDE 10
Related work
- Data science for social good
(predicting bike-share usage in Chicago’s Divvy bike system)
- Jake VanderPlas (modelling the effects of weather on bike
usage in Seattle)
SLIDE 11
Machine learning
SLIDE 12
Training set Machine learning algorithm Test set Learned estimator Predictions for test set
SLIDE 13
11 bikes
Training set
Tuesday 8:00 AM Downtown Sunny 0 bikes Tuesday 11:00 AM Downtown Sunny 2 bikes Tuesday 8:00 AM Downtown Rainy 2 bikes Tuesday 11:00 AM Downtown Sunny 1 bike Tuesday 1:00 PM Downtown Sunny
SLIDE 14
Test set
Tuesday 8:00 AM Downtown Sunny Tuesday 11:00 AM Downtown Sunny Tuesday 8:00 AM Downtown Sunny Tuesday 1:00 PM Downtown Sunny Tuesday 2:00 PM Downtown Sunny 11 bikes 1 bike 10 bikes 2 bikes 1 bike
SLIDE 15
Software overview
SLIDE 16
SLIDE 17
Libraries used
- Scikit-learn (machine learning algorithms)
- Pybikes (data collection) to collect data from the
Washington bike-share system
SLIDE 18
Machine learning algorithms
SLIDE 19
Decision Trees
SLIDE 20
Sunny Rainy Morning Morning Noon Noon 10,11 12,13 0,1 2,3 0,1 2,3 0,0 0,0
SLIDE 21
Random Forests
SLIDE 22
Random Forests
- Lots of decision trees
- Output given by the average of the output of all
trees in the forest
- Cannot overfit by adding more trees (note: RF can
- verfit on noisy datasets when there are too few
trees!)
SLIDE 23
Ada Boost
SLIDE 24
AdaBoost
- Analogy: student preparing for an exam in physics
- Topics covered: classical physics, thermodynamics,
electromagnetism, quantum physics
- They start by doing a practice exam
- They notice they didn’t do well on electromagnetism.
Ignore all other topics until they grasp electromagnetism.
- Do another practice exam
- Repeat… until it’s time for the exam
SLIDE 25
Thesis contribution
SLIDE 26
Data collection using Pybikes
SLIDE 27
Feature selection
SLIDE 28
SLIDE 29
SLIDE 30
SLIDE 31
Why is the “epoch” so important?
A missing time-related feature that has not been accounted for.
SLIDE 32
Genetic algorithms
- Hyperparameters - algorithm
configuration
- Can use GA to pick the “optimal” feature
set that provides the best prediction performance
- GAs did not improve the accuracy over
manually picked hyperparameters
SLIDE 33
Results
SLIDE 34
A customizable machine-learning package for predicting bike-share usage
SLIDE 35
SLIDE 36
SLIDE 37
SLIDE 38
Improvements on existing solutions?
SLIDE 39
Error metric: RMSE
SLIDE 40
Improvement
Poisson model (DSSG) Decision Tree Regressor Random Forest Regressor Ada Boost Regressor
1,75 3,5 5,25 7
Error (RMSE)
SLIDE 41
Further work
SLIDE 42
The vision
“I will be downtown at 8 am on Monday. Will the bike station be full?”
SLIDE 43
SLIDE 44
SLIDE 45
SLIDE 46